Saad [184] proposes to construct an incomplete LQ factorization
of a general sparse matrix. The idea is to orthogonalize the rows
of the matrix by a Gram-Schmidt process (note that in sparse matrices,
most rows are typically orthogonal already, so that standard Gram-Schmidt
may be not so bad as in general). Saad suggest dropping strategies for
the fill-in produced in the orthogonalization process. It turns out that
the resulting incomplete L factor can be viewed as the incomplete Cholesky
factor of the matrix 
. Experiments show that using 
in a CG
process for the normal equations: 
is effective for
some relevant problems.